Exa Web Research Monitor

Exa monitors are good at finding pages that never contain the words you searched for, which is the reason to run them and the reason a raw result feed is unreadable. This agent takes each run, throws out what is off brief or already covered, opens only the results that earn the cost of opening, and posts a single Slack digest saying why each survivor matters. What gets watched stays configured in Exa; what reaches your team is the judgment this adds on top.

What people use it for

• Research without the reading - Somebody on your team is already skimming forty links a week to stay current on a market. This does the skimming and hands over the four that changed something.
• Nothing arrives twice - Earlier runs are checked before anything is written, so a story that travels for three days across a dozen outlets reaches you once, from whichever source got there first.
• Papers your keywords would miss - Academic and engineering write ups routinely describe your problem using none of your vocabulary. Semantic monitoring catches them, and this layer decides which are worth a working day.
• The dropped count keeps you honest - Every digest ends with what was discarded and why. Read a fortnight of those lines and you know exactly where the brief is too loose, which no unexplained silence would ever tell you.

Before you fork

What has to exist in Exa before we fork this?

An Exa API key, at least one monitor already created there with its query and schedule, plus whichever Slack channel should carry the digests. The agent reads monitors and never creates them, so shaping the watch list stays a deliberate act you perform in Exa. Test Run stages a batch of results against your brief so you can read a digest before any of it is live.

What drives the cost of a run?

One agent run per batch of results, plus the Exa content calls, and content is the metered part. Because only the shortlist is opened, a tight brief is also the cheapest brief. A monitor returning forty items a night costs noticeably more than one returning five, so narrow the query in Exa before you widen the brief here.

How do we keep it from drifting off topic?

The brief is the whole control surface, and nothing else influences what is kept. If two weeks of digests show it letting adjacent material through, name that material in the brief as noise and it disappears from the next run. If it is dropping things you wanted, the dropped counts tell you which category to loosen.